The Godfather of AI, Geoffrey Hinton, was interviewed on 60 Minutes last night. He revealed that his original area of study was the human brain. He started by developing a digital model of the Neural Networks that humans use to process information. The fact that AI researchers have always tried to make AI just like us is a potential problem.
The video last night featured robots kicking a ball into a net. The point being made was that the robots had not been specifically programmed for that precise behavior. Rather the robots were “learning” in that actions that did not result in a goal were not reinforced, but actions that did result in the ball going into the net were.
According to Mr. Hinton it is a mistake to assume that AI is just “Auto Completing” or filling in the last word in a sentence. To do that the AI must “understand” everything leading up to that final task.
Mr. Hinton warns us that his brainchild could get away from us.
AI learns from us, and we are a hodgepodge of conflicting ideologies driven by an aggressive and competitive spirit. Do we really want AI to “learn” to be the same. Apparently so! One initiative in the field is to eliminate bias from the results AI produces. If that means that AI should draw from everyone’s thinking, that would be a mistake. AI needs to be limited to the factual and the verifiable.
While my wife and I were watching last night, my wife expressed the reasonable assumption that we will be able to control AI. My response was that we are not trying to control AI. We are trying to get ahead of everybody else in the use of it. The same mistake was made right after the atomic bomb was invented.